In 2026 Zhang, Gibeault and colleagues modelled spin networks with delayed interactions, motivated by coupled tunnel junctions, and report that "sufficiently long delays drive the steady-state probabilities toward equal state occupations even in strongly coupled systems". Their spins flip at a rate set by their own current state and their neighbours' delayed states. On a p-bit fabric, what a late read does turns on whether the update looks at the value the spin already holds. A heat-bath p-bit does not: it draws a fresh value from the field it reads. Redraw every spin that way on every tick, with every read d ticks old, and the fabric's frames split into d interleaved synchronous chains that never meet. Its law is the same at every delay: already wrong, as Lesson 16 began by showing, and no worse. Their rule does look, and two coupled spins under it agree less often as the delay grows. Lesson 16's repair for the shared clock pulls each spin toward its own current value, and that pull is the dependence a delay needs. Pinned at two, a coupled pair with no field sits nine thousandths from Boltzmann with reads one tick old and very nearly d times as far with reads d ticks old. Lesson 16's other repair fares no better: a spin waiting for its colour holds its value, and a coloured pair, exact with fresh reads, agrees only half the time with reads two ticks old. A fabric can win accuracy back by pinning harder, or by updating only when its reads are fresh. With reads three ticks old, either way kept about a third of the motion on the graphs measured; which was cheaper depended on the graph, by at most sixteen percent. When every wire has the same delay, a late read can change the law only through the value a spin holds now, and both of Lesson 16's repairs, pinning and colouring, add exactly that dependence.